3 Observations from a Bot that Chats with 50,000 People a Week
50,000 interactions in the first week and it hasn’t slowed down since.
A year ago I worked on a project where, in 6 months, our team planned, designed, tested, broke, fixed, broke again, fixed again, and finally deployed a bot that, in one week, interacted with 50,000 unique users. I won’t get into the bot itself nor where you can find it, but I will share a couple of observations that I think can only be made when you look at the data produced from 50,000 chats.
- People can be very polite to bots: You would be amazed how many people say thank you before they exit, or wish a bot good luck, say god bless or even compliment a bot. We gave our bot a personality, but it was still as surprise! Perhaps it’s demographics, perhaps the bot is really that good, but it never occurred to us that there were that many people out there who felt compelled to show kindness to a informational bot with a cheery personality.
2. People can be dicks to bots: Trolls are everywhere and they even take time to troll a bot. A fairly high percentage of people either troll bots or get pissed when the bot doesn’t respond correctly, road rage pissed.
We actually went back and enhanced the experience to respond appropriately to angry people and trolls, asking them to calm down or to stop bothering us — it actually worked pretty well.
3. People are very open to suggestion from a bot, but not always to the types of suggestions you might think: This isn’t something new we discovered with this project. In fact, in marketing and sales it’s quite well known that people are more open to suggestion in certain contexts than others. Our bot simply reinforced the importance of context and how it sometimes creates unexpected value that emerges from usage. Here’s a concrete example (not from our bot but analogous).
My wife signed up for a huge nationwide gym two months ago and did so via a chatbot with a very simple conversation tree that asked questions and populated basic information into a registration form.
She was asked about her fitness, sports she liked to play, what sorts of classes she was interested in. Buttons and yes/no questions guided most of the experience, but there were several opportunities to answer in free form text. The bot had decent guard rails, keeping the conversation flowing towards completing sign up. If she tried to talk about things the bot was not designed for, it would politely redirect the conversation. In 4 minutes she was done.
Then she went to LuLu lemon & spent $200 on work out clothes.
When is someone more likely to buy an umbrella? Just before it starts to rain. When might someone purchase $200 worth of work out gear? Within 30 minutes of signing up for a new gym. My wife talked to two entities that day and provided signals she was interested in stocking up on new work out gear.
Me and a bot.
Can you imagine if the owner of that chatbot could plausibly claim credit for the sale? Do you think my wife would have viewed a recommendation to purchase work out gear to be intrusive or obnoxious? Can you imagine if that chatbot signed 50,000 people up for new gyms a week?
Man alive is that a missed opportunity.
Perhaps the most important thing we’ve learned from our chatbot is this concept of emergent value. A good passage from a post on Emergence…
“As we continue to heighten connectivity and adopt analytical models that capture emergent properties, we may well see core business processes and conversational systems change dramatically. If we can observe, predict and react to user needs based on emergence, then we will need to rethink how conversational systems function at scale.”
At massive scale, patterns, trends and unique patterns emerge. Often these emergent behaviors are difficult to predict until actual scale is achieved and observed. I’m guessing at 50,000 weekly users, a fitness bot could sell a number of contextually relevant things other than gym clothes…
I think what’s missing and I’m not the first person to point this out, there needs to be a “bring your own machine learning model” platform that will enable an ecosystem where developers, brands, and users can easily connect to produce meaningful, value added outcomes for everyone. Where a gym signup bot suggesting a pair of LuLu Lemon pants to my wife before she even closes her chat bot window is no longer a dream.
Shameless plug coming right up!!!
Companies like cashbot.ai are working to help that gym signup bot (and all other sorts of bots) identify usage patterns from users behaviors, then in real time make product recommendations at just the right moment in the conversation. By doing so, a product recommendation can be a value add to the user rather than an irrelevant annoyance — monetization platform for chatbots and it’s 100% free.
Not a bad deal right!?!?
If you liked this post I’d really appreciate a clap or a share… It really helps us!
That’s it, I love you all!